What has changed
Allocator diligence has always had an unobserved first stage — the reading, asking around and searching that happens before a manager knows they are being considered. That stage is increasingly mediated by generative engines, because they compress a week of scattered reading into a paragraph.
The consequence is specific: a fund that does not appear in that paragraph is not rejected. It is never assessed. There is no meeting to lose, no data room declined, no feedback. The absence is invisible to the manager, which is what makes it dangerous.
Why funds are structurally disadvantaged here
- Marketing constraints mean many funds publish very little, so there is little for a system to retrieve and nothing to corroborate.
- Fund names are frequently ambiguous strings — several entities share similar names, and a system that cannot resolve which one you are will decline to name any.
- The most substantive material sits behind data-room walls, correctly. But that means the public record is thin relative to the operation it represents.
- Third-party databases and directories often carry stale entity information, and those sources are what gets retrieved.
The four mechanics
These are the same mechanics that decide visibility in any sector. What differs is the starting position, not the method.
- Retrieval access. If your robots.txt blocks the agents that read on behalf of these engines, nothing else matters. It is free to fix and it is the most common blocker we find.
- Entity resolution. Consistent naming, structured data matching the visible page, and corroborating profiles, so a system resolves you as one organisation rather than several similar strings.
- Passage independence. A section that only makes sense after reading two others cannot be lifted as an answer. Retrieval works at passage level.
- External corroboration. What independent sources say carries weight precisely because you do not control it. This is the slowest layer and the one competitors cannot buy quickly.
What this does not solve
Visibility is not allocation. Appearing in a generated answer puts a fund into a consideration set; track record, terms, operational diligence and relationships decide everything after that. Anyone selling AI visibility as a fundraising solution is overselling it.
And retrieval is non-deterministic. The same question produces different text and different citations between runs. A guaranteed citation should be treated as a warning about the supplier rather than a feature of the offer. What is measurable is answer share sampled repeatedly against a question set fixed before results are seen.
The compliance constraint is real
General-solicitation limits, performance-claim rules and jurisdiction-specific regimes shape what a fund may publish. None of the four mechanics above requires publishing performance data or soliciting. Entity clarity, accurate descriptions and corroborated factual information are not marketing claims.
Where the line sits is a question for your counsel, not for us. We will say plainly that we do not give regulatory advice, and we would be wary of any supplier who does.